iid tests for fmpe and score-based iid methods failing

[janfb]

The tests at test_vector_field_iid_inference have some open todos, xfail and skip cases which needs double checking and fixing.

sbi/tests/linearGaussian_vector_field_test.py

Lines 346 to 371 in e6f42a0

	# TODO: Currently, c2st is too high for FMPE (e.g., > 3 number of observations),
	# so some tests are skipped so far. This seems to be an issue with the
	# neural network architecture and can be addressed in PR #1501
	@pytest.mark.slow
	@pytest.mark.parametrize(
	"iid_method, num_trial",
	[
	pytest.param("fnpe", 3, id="fnpe-3trials"),
	pytest.param("gauss", 3, id="gauss-3trials"),
	pytest.param("auto_gauss", 8, id="auto_gauss-8trials"),
	pytest.param("auto_gauss", 16, id="auto_gauss-16trials"),
	pytest.param("jac_gauss", 8, id="jac_gauss-8trials"),
	pytest.param("jac_gauss", 16, id="jac_gauss-16trials"),
	],
	)
	def test_vector_field_iid_inference(
	vector_field_trained_model, iid_method, num_trial, vector_field_type, prior_type
	):
	"""
	Test whether NPSE and FMPE infers well a simple example with available ground truth.
	"""
	if vector_field_type == "fmpe":
	# TODO: Remove on merge
	pytest.xfail(reason="c2st to high, fixed in PR #1501/1544")
	num_samples = 1000

[janfb]

partly fixed in #1657 , duplicate of #1656